I’m sure we all wish that we could back up to locally-attached high-speed SSDs. But the reality is that many have to leave Time Machine to makes its hourly backups to networked storage. This inevitably results in poorer performance: even SATA/USB-C SSDs should achieve transfer speeds of around 500 MB/s, while the fastest that you’ll see over Gigabit Ethernet is a quarter of that, and that’s before the overheads of Time Machine and SMB.
In two previous articles I have looked at performance when backing up all-wirelessly in macOS 11.2.3 and using a combination of macOS 11.6 and Monterey beta. This article reports results using macOS 12.1 over Gigabit Ethernet.
With all-wireless 11.2.3, Time Machine took over 5 hours to back up 79 GB, giving an average transfer rate of around 4 MB/s. That improved with 11.6 to 7.9 MB/s. However, in the latter case, Finder copying of a single 10 GB file was significantly faster at almost 12 MB/s. As a result, I suggested that a useful rule of thumb is that TM backups are likely to proceed at around 0.67 of the rate of Finder copies.
For these tests, two M1 Macs were connected via a router using wired 1 Gbps Ethernet, with a maximum transfer rate of 125 MB/s. The client was an M1 Pro backing up from its internal SSD, with a read transfer rate of around 6,700 MB/s, and the server was an M1 Mac mini storing the backup to a sparse bundle on an external SATA/USB-C SSD, with a typical write transfer rate of about 500 MB/s.
Copying a 10 GB file from the M1 Pro to a shared APFS volume on the server achieved a transfer rate of 93.5 MB/s, which is 75% of the maximum achievable over the Ethernet connection between the Macs. Using my free app Stibium, average transfer speeds to the shared volume were 73 MB/s write, and 93 MB/s read, which are also reasonably close to the Ethernet maximum.
The first full backup over the network transferred a total of 61 GB in 287,960 files, giving an overall transfer rate of 35 MB/s. The following incremental backup transferred 15.1 GB in 585 files, at a transfer rate of 43 MB/s. These are 0.38 and 0.46 (respectively) of the Finder transfer rate, indicating a proportionally higher overhead than when using all-wireless connections (0.67), although their overall performance was 4-5 times faster than all-wireless.
Some users who back up to network storage have reported evidence of throttling. In macOS 12.1 backing up to shared storage on another Mac, indicated total CPU loads shown by Activity Monitor remained between 60-180% for much of the time. Most of the load was borne by the E cores of the client, while system processes also made good use of the four P cores in the first cluster, presumably to handle networking and file transfer. These are shown in the example CPU History window below.
There’s no evidence here of any throttling occurring in this test, and usage of E cores appeared similar whether backing up to local or network storage.
- As expected, Gigabit Ethernet wired connections resulted in much faster backups that all-wireless, by a factor of 4-5 times the transfer rate.
- However, wired network backups were slower than all-wireless, relative to the speed of Finder copies. This suggests that the network file protocol used (SMB) and/or Time Machine itself are reducing performance over faster network connections.
- When estimating expected backup performance over wired Ethernet connections, rather than using 0.67 times the Finder transfer rate (as expected for all-wireless backups), TM backups are likely to have about 0.4 of the performance of a Finder copy.
- If you need backup transfer rates higher than about 40 MB/s, Gigabit Ethernet is unlikely to deliver them. Faster network connections or local backup storage are obvious alternatives.
- For a typical incremental backup of up to 12 GB, wired Gigabit Ethernet connection to another Mac should complete within 5 minutes, and should therefore take an acceptable time. If your Time Machine backups exceed 12 GB each hour, then you’re likely to need a faster connection to the server, or automatic backups are likely to prove disruptively long.
- Third-party network storage, such as a NAS system, needs to achieve similar performance to compete with another Mac used as a server.